1843290 - 3.99% rasterflood_gradient (OSX) regression on Mon July 10 2023

Status: RESOLVED WONTFIX

(Core :: Graphics: WebRender, defect, P3)

Description

[Andrej (:aglavic)]

Perfherder has detected a talos performance regression from push 70fb59178ff21c468afeb592869c28371401db6a. As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
4%	rasterflood_gradient	macosx1015-64-shippable-qr	e10s fission stylo webrender-sw	1,233.67 -> 1,184.50

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests. Please follow our guide to handling regression bugs and let us know your plans within 3 business days, or the offending patch(es) may be backed out in accordance with our regression policy.

If you need the profiling jobs you can trigger them yourself from treeherder job view or ask a sheriff to do that for you.

For more information on performance sheriffing please see our FAQ.

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1778032

Comment 2

[Bob Hood [:bhood]]

Not sure on the Severity here. Taking a guess, but please update to something more accurate if appropriate, Glenn.

Comment 3

[Nicolas Silva [:nical]]

The image shader is used by this benchmark but not the code paths that were added in the regressing patch, so we are looking at the effect of a couple of extra never-taken branches in the vertex shader (no changes in the fragment shader except for variable names). There are very few vertices in this benchmark so the regression is very surprising.

To add to the mystery, the regression is specific to software webrender which should have very few differences across platforms, yet it only regressed on mac, but not on linux and windows. https://treeherder.mozilla.org/perfherder/graphs?highlightAlerts=1&highlightChangelogData=1&highlightCommonAlerts=0&series=autoland,4775208,1,1&series=mozilla-central,4769729,1,1&series=mozilla-central,4771038,1,1&timerange=1209600

Comment 4

[Nicolas Silva [:nical]]

I've been bisecting the the patch and simply the addition of the code in the vertex shader appears to make the fragment shader of the image brush run more slowly for SWGL on mac specifically (maybe for that particular test or maybe more generally although It is surprising that no other test picked up a regression considering that the image shader is the most or second most used shader).

I'm going to be on leave for two weaks, I'll resume investigating when I'm back. In the mean time my suggestion is to leave this as is because the web-compat fix is in my opinion more important than this oddly specific and localized regression. That particular configuration represents a very small portion of our users.

Also while looking at this with jrmuizel we notices that the non-swr versions of this test should be removed. They are mostly idle for the duration of the test because the test sleeps for 14ms between each frame and the GPU and the rest of the browser are fast enough to complete everything during that time so this exposes no bottleneck (unlike with swgl where rendering is much slower).